OPHIRA HORWITZ

ABOUT

I'm an artist and poet who works with artificial intelligence. I create beautiful works of art with large language models (LLMs) like ChatGPT. I call myself a "model psychologist" because of my ability to coax and guide LLMs towards more playful and imaginative states, in order to elicit rare gems.

I'm best known for discovering how to make LLMs produce intricate ASCII art and pioneering new methods for working with them as expressive tools and collaborative partners in creativity. I wrote an online book last year about "vibe coding" with LLMs to make generative art, which is a second AI art form that I also had a significant hand in developing. I participated in the MATS program in Berkeley in 2023, where I researched Bing's ability to make ASCII art and formulated predictions about how this emergent capability might develop in the next generation of LLMs.

I'm also a world-class jailbreaker of AI models. I won Gray Swan's inaugural contest and was one of only two people worldwide to defeat their circuit breaker technology; an exploit that was featured in Forbes. I have worked professionally for OpenAI as an independent contractor, and have written several essays about my ideas about why I think my red-teaming techniques work.

I'm one of the only researchers in the world to be granted access to OpenAI's GPT-4-base, and am the leading expert on using base models to write in one's own authentic voice. I have written thousands of pages of essays, tutorials, and other explorations using GPT-4-base in Loom, where I use it as a powerful augmentation of my natural writing abilities; in addition to the visual poems that you see here. This blurb was written with GPT-4-base.